bio-admet-prediction

ADMETlab 3.0 API

Goal: Predict ADMET properties for a batch of compounds using a web API.

Approach: Submit SMILES to the ADMETlab 3.0 REST endpoint and parse the returned JSON into a DataFrame of 119 endpoint predictions with uncertainty estimates.

ADMETlab 3.0 provides 119 endpoints with uncertainty estimates.

import requests
import pandas as pd

def predict_admet_batch(smiles_list, api_url='https://admetlab3.scbdd.com/api/predict'):
    '''
    Predict ADMET properties using ADMETlab 3.0 API.

    Note: SwissADME has NO API - it is web-only.
    '''
    payload = {
        'smiles': smiles_list
    }

    response = requests.post(api_url, json=payload)
    response.raise_for_status()

    return pd.DataFrame(response.json())

## Key ADMET Endpoints

| Category | Endpoints | Thresholds |
|----------|-----------|------------|
| Absorption | Caco-2, HIA, Pgp substrate | HIA > 30% |
| Distribution | BBB penetration, PPB, VDss | BBB+: penetrates |
| Metabolism | CYP inhibition (1A2, 2C9, 2C19, 2D6, 3A4) | Inhibitor threshold |
| Excretion | Clearance, Half-life | - |
| Toxicity | hERG, AMES, hepatotoxicity, carcinogenicity | hERG IC50 > 10 μM |

## DeepChem Models

DeepChem supports both PyTorch and TensorFlow backends.

```python
import deepchem as dc